On 10/24/2012 11:06 AM, Rick Hillegas wrote:
I'm wondering what we should tell users about the new
SYSCS_INVALIDATE_STORED_STATEMENTS procedure. Should we say:
I'd vote for 1 with a bit more of an explanation:
I) Run SYSCS_INVALIDATE_STORED_STATEMENTS whenever you think that your
metadata queries or triggers are mis-behaving, for example if they
throw a NoSuchMethod error on execution. Derby stores plans for triggers
and metadata queries in the database. These should be invalidated
automatically on upgrade and at other necessary times. Should you
encounter an instance where they are not, you have found a bug that you
should report, but one that you can likely work around by running
SYSCS_INVALIDATE_STORED_STATEMENTS.
I) Run SYSCS_INVALIDATE_STORED_STATEMENTS whenever you think that your
metadata queries or triggers are mis-behaving.
II) Run SYSCS_INVALIDATE_STORED_STATEMENTS whenever you upgrade the
version of Derby which you are using.
III) Only run SYSCS_INVALIDATE_STORED_STATEMENTS when tech supports
tells you to.
IV) Something else?
What follows are some additional musings which led me to ask this
question.
Thanks,
-Rick
---------------------------------------------------
My first step in buddy-testing the SYSCS_INVALIDATE_STORED_STATEMENTS
procedure was to read the user documentation in the Reference Manual.
The section on SYSCS_INVALIDATE_STORED_STATEMENTS talks about stored
prepared statements and mentions the SYSSTATEMENTS table.
I don't think that our doc set explains, anywhere, what a stored
prepared statement is. The Reference Manual section on SYSSTATEMENTS
doesn't explain what they are or why Derby creates them. I imagine
users must be a little confused by the explanation for
SYSTATEMENTS.USINGTEXT, which refers to the otherwise undocumented
CREATE STATEMENT and ALTER STATEMENT syntax. And I can't find any
other explanation of stored prepared statements in our doc set.
So when I read the section on SYSCS_INVALIDATE_STORED_STATEMENTS, I
imagined that a user might ask:
1) How do I know that I need to run this procedure?
2) Should I run this procedure every time I hard-upgrade the database?
3) Should I run this procedure whenever I upgrade the version of Derby
which I'm using?
I think that the following is true:
a) Derby persists stored prepared statements in order to improve the
performance of metadata queries and triggers the first time they run
after the database has booted. That is, stored prepared statements are
meant to save the first user the cost of compiling a metadata query or
trigger.
b) There are only 2 reasons for a user to forcibly recompile a stored
prepared statement:
i) Derby had a bug when it originally compiled the metadata query or
trigger and now that bug has been fixed.
ii) The persistent form of query plans has changed between versions
of Derby and the old metadata and trigger plans need to be discarded.
